Can you use yasa for sleep EEG analysis in mice?

[Fisher-Ding]

Is it possible to use the YASA library for sleep EEG analysis in mice?I already have markers for sleep staging.

[Fisher-Ding]

Because the sleep stages of mice are not the same as those of humans, only arousal, NREM, REM

[remrama]

The current YASA sleep staging algorithm was trained only on human PSG data, so it shouldn't be applied to mouse data. However there has been prior interest, and I believe some progress, on potentially adding mouse staging as a new YASA feature, though it has not been implemented yet. See #72.

Or do you mean you want to train a mouse model? (based on saying you have markers.) Either way, you should check out #72 and see if that answers your question.

[Fisher-Ding]

Here's the code I wrote using Quick Start Data (https://yasa-sleep.org/quickstart.html)

import mne
import pandas as pd
import yasa
import numpy as np
import matplotlib.pyplot as plt

Load the EDF file containing EEG data

raw = mne.io.read_raw_edf("/Users/ding/Desktop/yasa_example_night_young.edf", preload=True)
print("Original channel names:", raw.ch_names) # Display the original channel names

Drop unnecessary channels

try:
raw.drop_channels(["ROC-A1", "LOC-A2", "EMG1-EMG2", "EKG-R-EKG-L"])
except ValueError as e:
print(f"Error while dropping channels: {e}")
remaining_channels = raw.ch_names
print("Remaining channel names:", remaining_channels) # Display the remaining channel names

Get the original sampling frequency and resample to 100 Hz

original_sampling_rate = raw.info["sfreq"]
print("Original sampling rate:", original_sampling_rate) # Display the original sampling rate
raw.resample(100)
updated_sampling_rate = raw.info["sfreq"] # Update the sampling rate after resampling
print("Updated sampling rate:", updated_sampling_rate) # Display the updated sampling rate

Apply a band-pass filter (0.3 - 45 Hz)

raw.filter(0.3, 45)

Extract the data and convert it to microvolts (µV)

data = raw.get_data(units="uV") # data is a NumPy array
print("Data shape:", data.shape) # Display the shape of the data array

Load the sleep stage labels from a CSV file

hypnogram_file_path = "/Users/ding/Desktop/yasa_example_night_young_hypno.csv"

Read the CSV file, assuming the first column contains the sleep stage labels

sleep_stage_labels = pd.read_csv(hypnogram_file_path, header=None)[0] # Read the first column

Ensure the Hypnogram class from YASA is correctly imported

from yasa import Hypnogram

Create a Hypnogram object, assuming there are 3 sleep stages in the labels

hypnogram = Hypnogram(sleep_stage_labels, n_stages=3)

Visualize the hypnogram

yasa.plot_hypnogram(hypnogram)
plt.show()

Convert the hypnogram to integer values and then to a NumPy array

hypnogram_int_values = hypnogram.as_int()
hypnogram_np_array = hypnogram_int_values.to_numpy()

Upsample the hypnogram to match the resolution of the EEG data

try:
upsampled_hypnogram = yasa.hypno_upsample_to_data(
hypno=hypnogram_np_array, sf_hypno=1 / 30, data=raw
)
print("Length of upsampled hypnogram:", len(upsampled_hypnogram)) # Display the length of the upsampled hypnogram
except Exception as e:
print(f"Error during hypnogram upsampling: {e}")

Visualize the hypnogram again if needed

try:
yasa.plot_hypnogram(hypnogram)
plt.show()
except Exception as e:
print(f"Error while creating or plotting the hypnogram: {e}")

Plot the spectrogram for the 'C4-A1' channel with the hypnogram overlay

if "C4-A1" in remaining_channels:
try:
c4_a1_channel_index = remaining_channels.index("C4-A1")
yasa.plot_spectrogram(data[c4_a1_channel_index], updated_sampling_rate, hypno=upsampled_hypnogram)
plt.show()
except Exception as e:
print(f"Error while plotting the spectrogram: {e}")
else:
print("The channel 'C4-A1' is not present in the data.")

[remrama]

I'm actually confused how you got that. If you are using the yasa_example_night_young_hypno.csv file, that has 5 stages represented as integers. So the line where you create the Hypnogram with Hypnogram(sleep_stage_labels, n_stages=3) shouldn't work.

When I try that I get an error appropriately telling me that strings are required.

>>> import pandas as pd
>>> import yasa
>>> fp = yasa.fetch_sample("night_young_hypno.csv")
>>> stages = pd.read_csv(fp).squeeze()
>>> hyp = yasa.Hypnogram(stages, n_stages=3)

AssertionError: Since v0.7, YASA expects strings to represent sleep stages, e.g. ['WAKE', 'N1', ...]. Please refer to the documentation for more details.

But setting one up manually shows that the plotting works with 3 stages.

>>> import yasa
>>> import matplotlib.pyplot as plt
>>> stages = 10 * ["WAKE", "WAKE", "WAKE", "NREM", "NREM", "NREM", "REM", "REM", "REM"]
>>> hyp = yasa.Hypnogram(stages, n_stages=3)
>>> hyp.plot_hypnogram()
>>> plt.tight_layout()
>>> plt.show()

[Fisher-Ding]

I processed the yasa_example_night_young_hypno.csv file by changing 0 to WAKE,1 to REM, and 2,3,4 to NREM. all the Hypnograms that can be drawn for the three sleep stages

[Fisher-Ding]

Here's my code and dataset and what I'm trying to achieve, if you can help me out I'd greatly appreciate it!
#214

[remrama]

I see, thanks for the detailed follow-ups @Fisher-Ding. Indeed the YASA codebase needs to be modified to accomodate your usecase. The issue is with how the hypnogram is plotted within plot_spectrogram. I'm going to open an Issue for this and anyone is welcome to act on this asap. Thanks for pointing this out!

[remrama]

Let's move the discussion to #215.